Functional Memory

A hierarchical memory system that maintains outcome-sufficient context for critical enterprise decisions through unified memory-knowledge-reasoning integration

New to memory? Start with this overview, then explore Layered Architecture for implementation details, and User Model for dimensional framework design.

The Fundamental Memory Challenge

Traditional memory systems fail in mission-critical applications because they can't distinguish what information actually matters for outcomes. They either store everything (wasting resources) or use generic importance scoring (missing critical details). When making high-stakes decisions, this approach breaks down.

Amigo's memory system is designed to keep the outcome-sufficient user context (what we call L3) readily accessible during conversations. When a patient mentions chest tightness, the system can surface their heart condition history, anxiety patterns, and medication context without waiting for ad-hoc retrieval. This minimises latency while maintaining the information needed for real-time reasoning.

The result: healthcare decisions that properly account for how current symptoms connect to medical history, medication interactions, family patterns, and past treatment responses.

Critical functions need memory systems optimized for the use cases they serve, not for general performance benchmarks. The only important measure of the quality of a memory system is the statistical confidence the agent can achieve on memory-dependent tasks, particularly when supporting multi-dimensional success criteria that extend beyond technical accuracy to encompass social factors, confidence building, and organizational integration.

In enterprise contexts, this becomes especially critical when supporting complex decision-making processes that require comprehensive historical context and confidence-based reasoning across multiple dimensions of organizational success.

Amigo's Functional Memory System solves this by:

1. Maintaining L3 (the global user model) in active scope during live sessions so the agent can reason with the right interpretation depth while avoiding unnecessary retrieval churn

2. Creating multiple interconnected feedback loops between global patient understanding and local processing through professional identity-driven interpretation

3. Using net-new information accumulation where L3 determines both what constitutes genuinely new information and offers the interpretive lens for understanding all historical context

4. Implementing Boundary-Crossing Synthesis that prevents information density explosion while maintaining global context across processing boundaries when merging L2 episodic models into L3

Outcome-Sufficient Context Preservation

What "perfect" means: Memory maintains the complete set of information needed to make correct decisions-what statisticians call "sufficient statistics." Think of it like a medical chart that captures all clinically relevant data (allergies, conditions, medications) while omitting irrelevant details (what color shirt the patient wore). This isn't perfect recall of everything; it's perfect preservation of what matters for outcomes.

The Core Problem: Traditional memory systems fail because they can't determine:

• What information deserves perfect preservation

• How to maintain contextual relationships over time

• When to recontextualize information based on new understanding

Amigo's layered architecture solves this by maintaining high-fidelity associative binding between critical information and its context, operating as one of the six core components in our System Components orchestration framework. When you need vital facts, you get them with their complete context-every time-enabling confident decision-making within the Observable Problem -> Verification feedback cycle that characterizes reasoning-focused AI systems.

User Model: The Memory Blueprint

Framework Components

The user model is the functional blueprint that guides the entire memory system:

1. Dimensional Framework: Defines what information requires near-perfect preservation and the methodology to achieve it.

2. Memory Navigation: Guides and contextualize search to and reasoning over the important information and its proximal data.

3. Contextual Conditioning: Provides critical present snapshot context for interpretation or recontextualization of past information.

4. Information Gap Detection: Intelligently identifies what information is missing for the current real-time context.

Real-World Example:

When a patient reports "feeling stress in their leg after exercising," a generic system might simply search for similar phrases. Amigo's approach:

1. L3 global model consultation: Identifies past leg injury from user dimensions immediately available in memory

2. Contextualized understanding: Current complaint interpreted against complete injury history without retrieval

3. Professional identity filtering: Physical therapy context shapes clinical interpretation priorities

4. Temporal pattern recognition: Distinguishes between temporary pain and chronic condition progression

This allows the system to provide responses that account for the full context-something generic memory systems fundamentally can't do.

Layered Memory Architecture

1

L0 Raw Transcripts

Complete conversation records that serve as ground truth for historical recontextualization during rare live session expansions and as source material for post-processing extraction.

2

L1 Extracted Memories

Net-new information accumulated through extraction with L3 anchoring, where L3 determines what's genuinely new and offers interpretive lens from complete historical perspective.

3

L2 Episodic User Models

Synthesized understanding from extracted memories with L3 anchoring, maintaining coherence across processing boundaries while preventing information density explosion.

4

L3 Global User Model

Complete merged understanding across all time that remains constantly in memory during live sessions, providing immediate access to all functionally important dimensions with professional identity-driven interpretation.

Key Features of Amigo's Memory System

1. Recent Information Guarantee

Amigo guarantees that recent information (last n sessions based on information decay for use case) is always available for:

• Full reasoning over the complete context

• Perfect recall of all details

• Recontextualization based on new understanding

This solves the fundamental problem of information decay that plagues traditional systems.

2. Rare Recontextualization Mechanism

When live session expansion is needed (rare, adds latency), Amigo employs a dual anchoring mechanism:

1. Memory-Knowledge-Reasoning Integration: L3 supplies memory at the right interpretation depth for knowledge application and reasoning without retrieval latency, making expansion rare because both the contextual foundation and immediate availability needed for clinical reasoning are already present

2. Gap-Specific Retrieval: Only retrieves missing gaps rather than broad searches, with queries written against L3 enabling deeper insight extraction

3. Recontextualized Understanding: Past L0 conversations recontextualized against current L3 understanding, enabling reasoning beyond simple retrieval

4. Professional Context Filtering: Service provider background guides what constitutes meaningful gaps requiring historical expansion

5. Temporal Synthesis: L3 bridges live session context with historical L0 context through dual anchoring mechanism

3. Dimensional Evolution and Clinical Intelligence

Unlike traditional systems that struggle with changing information, Amigo creates functional clinical intelligence through:

Core Capabilities:

1. Professional identity guiding interpretation at every level of the memory hierarchy

2. System evolution of attention patterns based on discovered patient patterns

3. Adaptive Dimensional Optimization: When the system detects drift between user dimension definitions and optimal interpretation patterns for a patient group, it can modify dimensional definitions and perform complete temporal backfill

Advanced Features:

1. Replay-Based Reinterpretation: Data backfill operates like a replay system, regenerating memory extraction (L0->L1), episodic user model synthesis (L1->L2), and L3 evolution (L2->L3) across all historical time with the superior dimensional framework

2. Group-Level Intelligence: This enables reinterpretation of entire patient cohorts with optimal information interpretation, depth, granularity, and angle based on discovered clinical patterns

3. Clinical Outcome Optimization: As understanding evolves, dimensional definitions can be updated with system backfilling by recomputing interpretations based on new dimensional understanding, improving safety, patient experience, and clinical outcomes

4. Continuous Knowledge Flow: Multiple interconnected feedback loops between global (L3) and local (L0/L1) processing ensure no information loss at processing boundaries

Concrete Example: Discovering Hidden Patterns

Consider a patient whose blood sugar control seems randomly unstable:

• Week 1-4 (L1 extraction): System captures seemingly unrelated mentions-work deadlines Tuesday, feeling stressed Thursday, missed medication Friday. Each seems like noise.

• Month 2-3 (L2 accumulation): Patterns emerge from accumulated L1 data. A 2-3 week cycle appears: work stress -> medication timing disruption -> blood sugar instability.

• Quarter 1-3 (L3 cross-episode analysis): Comparing multiple quarterly episodes reveals this isn't random-it's a stable functional dimension. The stress-medication-timing interaction becomes part of L3's dimensional blueprint.

• Result: What looked like random instability is actually a discoverable pattern. Now the system can proactively intervene when work stress patterns emerge, preventing blood sugar episodes.

This discovery was only possible because:

1. L1 captured seemingly irrelevant details (unfiltered extraction)

2. L2 aggregated over sufficient time for patterns to emerge (temporal aggregation)

3. L3 identified the pattern across multiple episodes (cross-episode analysis)

At population scale, only 10-50 such functional dimensions typically explain substantial outcome variance. The sparsity isn't imposed-it emerges as true causal patterns become visible while noise averages out.

4. Enterprise Customizability

Amigo's memory architecture is fully customizable for enterprise-specific needs through a comprehensive implementation process that our Agent Engineers will work with you on.

1. Critical Function Assessment: Identify functions requiring near-perfect memory and map critical information types & hierarchy based on your use cases.

2. Memory Design: Configure memory topology and define user dimensions + parameters.

3. Integration & Deployment: Deploy memory system, connect to existing data sources and initialize user models.

4. Verification & Optimization: Validate functional performance, optimize dimensional parameters to increase performance where necessary.

Memory in the Unified Cognitive System

We've covered what the memory system does and how it works. To understand why this architecture matters, we need to see how memory integrates with the broader Amigo system. Memory doesn't operate in isolation-it's one component of a unified cognitive architecture where multiple systems work together to enable clinical intelligence.

Unified Context for Decisions

The system assembles context C from multiple sources:

• Context Graphs define problem structure (what kind of clinical interaction is this?)

• Professional Identity provides interpretation priors (what matters to a physician vs. physical therapist?)

• Functional Memory maintains sufficient statistics (what do we know about this patient?)

• Constraints ensure safety limits

• Evaluations define success criteria

L3 provides the functional dimensions that, combined with professional identity and problem structure, form the complete context needed for clinical decisions.

How Memory Enables System Evolution

The hierarchical memory architecture creates a self-improving system through the macro-design loop:

Better Models -> Better Problem Definitions -> Better Verification -> Better Models

Without hierarchical memory maintaining sufficient statistics across timescales:

• Each interaction would start from scratch

• Patterns wouldn't accumulate into understanding

• Population-level learning would be impossible

• Long-horizon problems (tracking patient progress over months) would remain intractable

With memory preserving outcome-relevant patterns at multiple timescales:

• L1 captures what's new in each interaction

• L2 accumulates patterns over weeks/months

• L3 maintains stable functional dimensions discovered across episodes

• Backfill enables reinterpretation when understanding evolves

This compound loop is what transforms individual interactions into organizational intelligence. It's why memory isn't just storage-it's the foundation for a system that gets better over time.

Clinical Intelligence Through Memory-Knowledge-Reasoning Integration

Amigo achieves functional clinical intelligence by recognizing that memory, knowledge, and reasoning are not isolated functions but deeply intertwined facets of a single cognitive problem. L3 being constantly in memory provides the right interpretation, precision, and depth needed to power effective knowledge application and reasoning:

1. Complete Memory-Knowledge-Reasoning Integration: L3 provides memory at the precise interpretation depth required for clinical knowledge application with immediate availability, enabling reasoning that operates on complete contextualized information

2. Unified Context Foundation: L3 ensures complete unified context across memory, knowledge, and reasoning, where high-quality recontextualization emerges from having complete patient understanding immediately available for knowledge synthesis

3. Perfect Interpretive Depth: Memory is maintained at the exact precision and granularity levels needed for all reasoning tasks with immediate access-clinical decision-making gets the contextual depth it requires, care coordination gets what it needs, all without retrieval delays

This creates comprehensive contextual awareness essential for medical intelligence performance, where healthcare decisions require understanding how current symptoms connect to established patterns, medication interactions, family history, and treatment responses.

High-Bandwidth Cross-Layer Integration

Integration Pathways

Amigo achieves functional clinical intelligence through sophisticated high-bandwidth integrations between information hierarchies:

L3 <-> L0 Direct Integration

L3 provides interpretive context for direct L0 access, serving as a temporal bridge between present understanding and raw historical events, ensuring historical data is interpreted through complete current patient context.

L3 <-> L1 Extraction Guidance

Every L0->L1 extraction operates with complete awareness of the existing L3 global snapshot, ensuring new information is extracted in proper context rather than as disconnected fragments. The current L3 global snapshot feeds into extraction, preventing isolated session misinterpretations and ensuring continuous global (L3) to local (L0/L1) and local-to-global knowledge flow.

User Understanding <-> Dimension Definition Feedback Loops

The system creates nested feedback loops with object level (direct clinical application), meta level (dimension definition evolution based on pattern recognition), and meta-meta level (framework optimization based on meta-analysis of dimensional evolution patterns).

Optimization Impact

Memory as Safety Foundation

The Functional Memory System serves as a critical safety mechanism within Amigo's comprehensive safety framework. By striving for perfect recall of safety-critical information through L3's constant availability, the system ensures that safety decisions always consider complete context with proper clinical interpretation.

This manifests in several ways:

• Crisis Prevention: Past crisis indicators and risk factors remain immediately accessible, enabling proactive intervention

• Medication Safety: Complete medication history and adverse reactions guide all pharmaceutical discussions

• Risk Awareness: L3's dimensional framework prioritizes safety-relevant information with "perfect" precision requirements

• Safe Recontextualization: The dual anchoring mechanism ensures historical events are understood through current safety understanding

As detailed in Operational Safety, this memory-safety integration means protection emerges naturally from the same cognitive processes that drive all system behavior, rather than requiring separate safety filters that could be bypassed or fail.

Conclusion: Memory That Serves Patient Function

Critical Healthcare Reality: In critical healthcare contexts, memory that works "most of the time" is memory that doesn't work at all.

Patient safety requires memory systems that deliver:

• Perfect recall of critical information when patient safety depends on it

• Complete preservation of vital context across provider transitions and time

• Efficient identification of information gaps before they impact care decisions

• Understanding of information evolution over time as patient conditions change

The Amigo Advantage

Amigo's Functional Memory System delivers complete reliability through L3 being constantly available during patient interactions. The system provides everything needed to serve the patient with immediate access to complete context at the right interpretation depth, enabling clinical decision-making with full contextual awareness and zero retrieval latency that would degrade reasoning quality.

For medical functions where failure isn't an option, Amigo provides memory that works when patients need it most.